I usually agree with the EFF on things, but after reading their linked https://www.eff.org/deeplinks/2025/05/us-copyright-offices-d... I couldn't disagree more. An LLM is a predict the next word algorithm. If the model is overfitting, it's basically copy paste. There have been several documented instances where that happened and full GPL code, including headers and attribution were copy/pasted by the "AI."

AI is essentially copy paste with more steps. The part that AI companies use to defend this is ?how are we supposed to decide how much each author deserves? They try to wave this away, but their own model can tell them. Their models work off of weights. They can determine how much each work contributed based on those weights, so it's dishonest for them to argue it isn't possible. The way the models are engineered now don't make this possible, but that's intentional and we can all recognize that. They throw up their hands and claim it's not possible because they simply don't want to pay.

The most infurating thing however is how AI companies sidestep the IP rights of authors, but then claim to own those IP rights when their own generated output leaks. Anthropic filed DMCA takedowns on the leaked claude code repos, claiming ownership over something they explicitly have stated is almost entirely AI generated as part of their marketing. They take code, mix it up just enough to scrub away the GPL or whatever license belongs on it, then try to claim ownership of the result, in spite of the Copyright Office repeatedly stating that AI generated works have no copyright protection at all.

Agreed. Moreover, the authors of copyright law could never have anticipated this type and scale of abuse. Maybe the companies are legally in the right, maybe not, but that's irrelevant for the question of whether it's ethical. The EFF's post definitely goes against their mission to "ensure that technology supports freedom, justice, and innovation for all people of the world."